Sorting of Similar Shaped Objects Using Object Detection Method

To detect only specific similar shaped objects with YOLO, the paper proposes a series of procedures as an automatic detection method. The method is applied to quality control of precious paint products. Paints are produced through the milling process of powders. Tiny powders randomly attach to larger powders, called asteroid powders. These are degrading the quality of the product. The asteroid powders should be detected automatically by other powders similar in shape to them. In the paper, various YOLO models which differ in size are constructed by expanding the number of evaluators and data. YOLO models output class names and coordinate information. It needs to be optimized to evaluate performance. The method optimized some information and evaluated the performance of the models through validation. The paper discusses the evaluation characteristics of the evaluator and the model. In addition, it provides feedback to the evaluator, making the prediction results visible and understandable to the evaluator.